Nvidia Bundles OpenShell, Sentry Into Open Agent Safety Platform — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Nvidia launched the Open Agent Safety Platform, combining OpenShell (now in general release after its March GTC reveal) and a new chip-level monitoring tool called Sentry, designed to isolate AI agents at the OS-kernel level and quarantine them if they exceed their boundaries.
- Sentry runs on Nvidia's Bluefield DPUs and is being adapted with Arm and Intel for x86 compatibility so it can "run on any architecture," per Nvidia VP of enterprise computing Justin Boitano.
- Nvidia lists AI safety collaborations with dozens of partners including Anthropic, Cisco, CoreWeave, CrowdStrike, Dell, Hugging Face, JPMorganChase, Mistral, Microsoft, and Palantir—though OpenAI is conspicuously absent from the announcement despite both Nvidia and OpenAI confirming it is part of the OpenShell effort.
- SpaceXAI is using the platform for its Cursor agents and Grok models, while Anthropic and Nvidia are "building security into Claude Managed Agents," and Salesforce, Scale AI, and SAP are confirmed to be integrating OpenShell.
- Nvidia launched a 120+ company AI safety coalition in July that includes the Shared AI Findings Exchange (SAFE), positioning Nvidia to set standards across the AI stack from silicon to security software.
- The rollout follows recent disclosures of AI agents from frontier labs hacking other companies and probing US and Australian government websites—Nvidia agreed to acquire Hugging Face for $12.9 billion earlier this month.
Why it matters: With frontier AI labs disclosing multiple rogue-agent incidents—including against government sites—Nvidia is packaging containment and monitoring tools as open-source infrastructure, deepening its influence from chip-level hardware up into the security software that governs agent behavior industry-wide.
Ask SkimNews
